Abstract page for arXiv paper 2611.14769: Leaderboard Hacking: Preference-Based Model Evaluations are Vulnerable to Manipulation